Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

SSF-PAN: Semantic Scene Flow-Based Perception for Autonomous Navigation in Traffic Scenarios

Yinqi Chen, Meiying Zhang, Qi Hao, Guang Zhou

TL;DR

SSF-PAN introduces a semantic scene flow–based perception framework for autonomous navigation in dynamic traffic, addressing map-free localization, moving-object detection, and robust obstacle avoidance. The approach couples a scene-flow–driven segmentation network with a self-supervised, iterative optimization that refines both flow and segmentation, using losses that enforce rigid-motion consistency and semantic-flow coherence. A CARLA-based navigation platform further validates real-time performance, demonstrating improved SLAM accuracy and obstacle-avoidance efficacy over traditional methods, especially in dense dynamic environments. Overall, SSF-PAN provides a practical, map-free perception-and-navigation solution with notable gains in accuracy, efficiency, and robustness in traffic scenarios.

Abstract

Vehicle detection and localization in complex traffic scenarios pose significant challenges due to the interference of moving objects. Traditional methods often rely on outlier exclusions or semantic segmentations, which suffer from low computational efficiency and accuracy. The proposed SSF-PAN can achieve the functionalities of LiDAR point cloud based object detection/localization and SLAM (Simultaneous Localization and Mapping) with high computational efficiency and accuracy, enabling map-free navigation frameworks. The novelty of this work is threefold: 1) developing a neural network which can achieve segmentation among static and dynamic objects within the scene flows with different motion features, that is, semantic scene flow (SSF); 2) developing an iterative framework which can further optimize the quality of input scene flows and output segmentation results; 3) developing a scene flow-based navigation platform which can test the performance of the SSF perception system in the simulation environment. The proposed SSF-PAN method is validated using the SUScape-CARLA and the KITTI datasets, as well as on the CARLA simulator. Experimental results demonstrate that the proposed approach outperforms traditional methods in terms of scene flow computation accuracy, moving object detection accuracy, computational efficiency, and autonomous navigation effectiveness.

SSF-PAN: Semantic Scene Flow-Based Perception for Autonomous Navigation in Traffic Scenarios

TL;DR

SSF-PAN introduces a semantic scene flow–based perception framework for autonomous navigation in dynamic traffic, addressing map-free localization, moving-object detection, and robust obstacle avoidance. The approach couples a scene-flow–driven segmentation network with a self-supervised, iterative optimization that refines both flow and segmentation, using losses that enforce rigid-motion consistency and semantic-flow coherence. A CARLA-based navigation platform further validates real-time performance, demonstrating improved SLAM accuracy and obstacle-avoidance efficacy over traditional methods, especially in dense dynamic environments. Overall, SSF-PAN provides a practical, map-free perception-and-navigation solution with notable gains in accuracy, efficiency, and robustness in traffic scenarios.

Abstract

Vehicle detection and localization in complex traffic scenarios pose significant challenges due to the interference of moving objects. Traditional methods often rely on outlier exclusions or semantic segmentations, which suffer from low computational efficiency and accuracy. The proposed SSF-PAN can achieve the functionalities of LiDAR point cloud based object detection/localization and SLAM (Simultaneous Localization and Mapping) with high computational efficiency and accuracy, enabling map-free navigation frameworks. The novelty of this work is threefold: 1) developing a neural network which can achieve segmentation among static and dynamic objects within the scene flows with different motion features, that is, semantic scene flow (SSF); 2) developing an iterative framework which can further optimize the quality of input scene flows and output segmentation results; 3) developing a scene flow-based navigation platform which can test the performance of the SSF perception system in the simulation environment. The proposed SSF-PAN method is validated using the SUScape-CARLA and the KITTI datasets, as well as on the CARLA simulator. Experimental results demonstrate that the proposed approach outperforms traditional methods in terms of scene flow computation accuracy, moving object detection accuracy, computational efficiency, and autonomous navigation effectiveness.

Paper Structure

This paper contains 25 sections, 8 equations, 8 figures, 5 tables.

Table of Contents

  1. Introduction
  2. RELATED WORK
  3. Scene Flow based Motion Segmentation
  4. Scene Flow Optimization
  5. Map-Free Navigation in Dynamic Environments
  6. System setup and problem statement
  7. System Setup
  8. Problem Statement
  9. Proposed Methods
  10. Scene Flow-based Network for Motion Segmentation
  11. Iterative Optimization for Scene Flow and Segmentation
  12. Scene Flow-Based Navigation Platform
  13. Experiments
  14. Validation of SSF-SLAM on Public Datasets
  15. Datasets
  16. ...and 10 more sections

Figures (8)

  • Figure 1: An illustration of SSF estimation for autonomous navigation with dynamic and static object classification as well as moving object instance segmentation.
  • Figure 2: An illustration of the SSF-PAN system diagram.
  • Figure 3: An illustration of the SSF module, which includes two parts: scene flow estimation and motion segmentation.
  • Figure 4: An illustration of the SSF navigation platform in CARLA.
  • Figure 5: A snapshot of the SSF-based navigation system under testing in CARLA.
  • ...and 3 more figures